Community Co-Designed AI

The PRoblem

Building Hyper Local AI

Small coastal communities are often missing from national datasets and service directories. In Martham, for example, the nearest Alzheimer's Society office is over two hours away by public transport. Yet existing local groups and activities could already help people living with dementia.

That gap matters for social care. A care coordinator building a plan that fits the person and the place has difficulties using a basic search as information about social care is disparate.

AI search tools don't fix this on their own: research shows they too describe places more like a tourist than a local (Gao and Thebault-Spieker, 2026). But improving them to be hyper-local presents ethical issues.

The question: how might we co-create a community app in a way that is ethical and respectful of small coastal communities?

My Role

User researcher and responsible AI advocate

I was responsible for exploring how AI being developed for hyper-local searching should be developed. In practice, that meant:

collating initial ARISE team insights on local resources, capacities and the relations between them into a knowledge graph schema
designing and facilitating co-design sessions with ARISE's Community of Practice and local community groups
codesigning with local community's the quality assurance instructions for the final tool, such as "never invent details about a village"
facilitating discussions on AI literacy and information bias

I work with my line manager.

Approach

Study and think

I built a preliminary knowledge graph from ARISE's resilience map, its survey of community resources and a review of others' work.

Prepare two things a community can shape

A knowledge graph, or "mind map" of how local places, groups, activities and organisations connect and

instructions, or set rules for how the tool answers. I prepared generic versions of both as starting points.

Build in safeguards from the start

The prototype runs on a university-managed device in a closed system. It's framed as a research prototype for discussion, not a service being rolled out.

Test and co-create with the Community of Practice

Members reviewed a generic knowledge graph of village resources in three groups: resources, capacities and relations. They marked what was missing, wrong, too sensitive to include or extremely important.

Next: village workshops, then share

Residents will shape their own knowledge graph and instructions. We'll then share the prototype with councils, care coordinators and policymakers to ask whether and how it should be used.